Excel AI Step-by-Step: Install It, Use the Interface, and Write Your First Prompts

Now that you understand the ideas, this tutorial gets you set up. You'll install AI for Excel, learn your way around the interface, and write your first real prompts.

Table of Contents

• Installing AI for Excel
• Getting Around the Interface
• Your First AI Prompt
• A Worked Example
• Common Problems
• Practice

Installing AI for Excel

Copilot in Excel

• Needs Microsoft 365 (a subscription, not the free web version) plus a Copilot license (Copilot Pro or Microsoft 365 Copilot).
• If you have both, a Copilot button appears in the Home ribbon.
• No button, no license — that's the first thing to check.

Excel Labs (free alternative)

1. Go to Insert → Get Add-ins.
2. Search for Excel Labs and install it.
3. Use the LABS.GENERATIVEAI function from any cell. It takes a prompt and an API key:

=LABS.GENERATIVEAI("Write a VLOOKUP to find price by product name", OPENAI_API_KEY)

Python in Excel (optional)

Excel can also run Python code natively with the =PY(...) function. It's powerful but not required for AI basics — skip it until you're comfortable with prompts.

Getting Around the Interface

When you open Copilot, a panel appears on the right side of Excel.

• Text box at the top — where you type your request in plain English.
• Suggestion chips below it — one-click prompts like "Show insights" or "Generate a chart." Clicking one runs it instantly.
• Copilot works on the selected range. If your sheet has several tables, mention which one you mean.

The interface is just two things: a prompt box and an answer area. Everything else in Excel works exactly as before.

Your First AI Prompt

There's one formula for good prompts:

What you want + which data + what you want back.

Start simple. These all work on a normal sales sheet:

"What's the total sales in this table?"

"Show me the average order value by region."

"Create a pivot table of sales by product category."

"Explain why column D has #VALUE! errors."

Two habits that save time:

• Be specific about the data. Say "use column C, not B" if the first answer uses the wrong column.
• Review the formula before trusting it. AI writes it, but you approve it. Read the result, don't just accept it.

If the answer is wrong, don't rephrase wildly. Say what's off: "That included empty rows — only count rows with data." Refinement is normal, even for experienced users.

Prompt templates to copy. These patterns cover most day-to-day requests:

A Worked Example

Theory is one thing — here's a full run. Say your sheet looks like this:

Prompt 1 — understand your data.

"What is the total revenue per product?"

AI multiplies Units × Unit Price and groups by product. Expected answer:

Prompt 2 — drill down.

"Which region sold more units?"

North: 57, South: 47. North wins.

Prompt 3 — visualize.

"Create a bar chart of revenue by product."

AI inserts the chart for you. Done — the whole analysis took three prompts.

The point: you never typed a single formula. But you did check the numbers — Laptop 16,000 matches 12×800 + 8×800 — which is exactly how you should treat every AI answer.

Common Problems

Practice

Try these on a small sheet you already have:

1. Ask for a summary of your data.
2. Ask for one formula you'd normally look up.
3. Ask for a chart you've never made before.
4. Break the AI: give a vague prompt, then fix it with a specific follow-up.